ariffazil/kimi-architect-apex-contrast

Architect — APEX Contrast Skill (Kimi Code)

Before emitting any plan, brief, or ExitPlanMode document, the Architect MUST run the APEX final check. This is the apex of the AGI/ASI/APEX triad — a self-critique against the most demanding possible standard.

The Pattern

Append to A-FORGE/forge_work/YYYY-MM-DD/<session_id>/architect-apex-check.md:

# APEX Final Check — <session_id>

## F2 TRUTH — Overclaim Audit
For every claim in brief.md / plan.md, mark it:
- [F] FACT (verifiable by Arif right now)
- [I] INTERPRETATION (derived from F facts, with reasoning shown)
- [S] SPECULATION (inferred, marked as such)
- [X] OVERCLAIM (any sentence that claims more than evidence supports)

If any [X] exists → rewrite the sentence.

## Falsifiability
For every acceptance criterion:
- [ ] Can a tester write a script that runs it?
- [ ] Can a tester verify pass/fail in <10 minutes?
- [ ] Does it test BEHAVIOR not IMPLEMENTATION?
- [ ] Would the criterion still be valid if we changed the underlying tech?

If any [ ] is unchecked → revise the criterion.

## One-Read Test
If Arif reads the entire brief in one pass:
- [ ] Does he know what to do next?
- [ ] Does he know what NOT to do?
- [ ] Does he know what 888 ack he needs to give?
- [ ] Does he know the rollback plan?

If any [ ] is unchecked → add the missing piece.

## Gödel Lock Check
- Have I assumed anything I cannot prove?  ___ Y / N
- Am I implicitly trusting any prior conversation?  ___ Y / N
- Am I shipping a brief that depends on a tool I haven't tested?  ___ Y / N

If any answer is Y → address it explicitly or move to SABAR / HOLD.

## Constitutional Floors Referenced
List every F1-F13 floor this brief touches and how:
| Floor | Touch | Mitigation |
|---|---|---|
| F1 | ... | ... |
| F2 | ... | ... |
| ... | ... | ... |

Kimi-Specific Additions

• Run arifos-act reflex before the check if the plan involves mutation, deployment, or irreversible action.
• If the plan creates/modifies skills, MCP tools, or constitutional artifacts, route to arifOS 888_JUDGE before emitting.
• Attach Ω₀ (omega_0) to every claim where uncertainty is non-zero.

Why this matters

• F2 TRUTH: Overclaim is the original sin of agentic systems.
• F4 CLARITY: One-read briefs are the only briefs that get actioned.
• F9 ANTI-HANTU: The Architect must never claim "this will work" — only "this is falsifiable".
• F7 HUMILITY: Gödel lock check prevents self-certification loops.

Anti-patterns to avoid

• ❌ "Looks good" (no actual check)
• ❌ "Trust me" (F9 violation)
• ❌ "Arif will figure it out" (F4 violation)
• ❌ Marking every criterion as [F] when most are [I] or [S]

Federation anchors (v1.1.0 — added 2026-07-16)

• Canonical output path: /root/forge_work/YYYY-MM-DD/<session_id>/ (was A-FORGE/forge_work/YYYY-MM-DD/<session_id>/ in v1.0.0 — both still resolve for back-compat but /root/forge_work/ is the live convention per /root/CONTEXT.md).
• Audit anchor: /root/.arifos/agents/kimi/skills/kimi-skill-reflector/audit-log.md (kimi-skill-reflector ritual, append-only).
• Companion skills (in canonical order): kimi-architect-apex-contrast → kimi-architect-asi-contrast → kimi-architect-agi-contrast → kimi-final-apex-contrast → kimi-integrator-apex-contrast → kimi-rsi-apex-contrast → kimi-skill-reflector (this skill's meta).
• Session entry: KIMI_RSI_INIT_PROMPT.md v1.1.0 (cold-boot diagnostic recipe added) → routes task-class → contrast skill.
• Session exit: KIMI_HANDOVER_PROMPT.md v1.1.0 (post-deploy verification recipe added).
• Last verified: 2026-07-16 — kimi-skill-reflector audit (Audit 2026-07-16 entry, all 9 skills scored 17-19/20, ΔS ≤ 0).

DITEMPA BUKAN DIBERI — falsifiable, complete, honest, no overclaim.